The design and research of an offline voice interaction system for eVTOL (Electric Vertical Takeoff and Landing) aircraft consider the actual operating scenarios of low-altitude aerial vehicles, especially the complex environment of urban low-altitude flight. Addressing key issues such as high pilot operational complexity, information overload, low interaction efficiency, and poor real-time network stability, this study focuses on developing a real-time and efficient offline voice interaction system to achieve natural human–machine communication. The research involves the exploration of a self-supervised learning pre-training framework that combines the Whisper model with Large Language Models (LLMs), fine-tuning the model for low-altitude applications, and completing the design of the voice interaction system. It achieves a breakthrough in high-accuracy offline speech recognition and response. The results of actual deployment validation indicate that the system possesses capabilities such as offline natural language processing, high-accuracy speech recognition, and real-time response, meeting the needs of low-altitude eVTOL flight and providing strong assurance for improving operational efficiency and safety in low-altitude flying.